Professional Documents
Culture Documents
Textbook Longitudinal Multivariate Psychology Emilio Ferrer Ebook All Chapter PDF
Textbook Longitudinal Multivariate Psychology Emilio Ferrer Ebook All Chapter PDF
Emilio Ferrer
Visit to download the full and correct content document:
https://textbookfull.com/product/longitudinal-multivariate-psychology-emilio-ferrer/
More products digital (pdf, epub, mobi) instant
download maybe you interests ...
https://textbookfull.com/product/fundamentals-of-satellite-
remote-sensing-emilio-chuvieco/
https://textbookfull.com/product/multivariate-data-analysis-
joseph-f-hair/
https://textbookfull.com/product/practical-multivariate-
analysis-6th-edition-afifi/
https://textbookfull.com/product/practical-multivariate-
analysis-6th-edition-afifi-2/
Problems in Quantum Mechanics with Solutions 2nd
Edition Emilio D'Emilio
https://textbookfull.com/product/problems-in-quantum-mechanics-
with-solutions-2nd-edition-emilio-demilio/
https://textbookfull.com/product/actions-and-invariants-of-
algebraic-groups-second-edition-ferrer-santos/
https://textbookfull.com/product/longitudinal-data-analysis-
autoregressive-linear-mixed-effects-models-ikuko-funatogawa/
https://textbookfull.com/product/multivariate-analysis-for-the-
behavioral-sciences-kimmo-vehkalahti/
https://textbookfull.com/product/financial-markets-theory-
equilibrium-efficiency-and-information-2nd-edition-emilio-
barucci/
LONGITUDINAL MULTIVARIATE
PSYCHOLOGY
Sponsored by the Society of Multivariate Experimental Psychology, the goal of this series
is to apply statistical methods to significant social or behavioral issues, in such a way so as to
be accessible to a nontechnical-oriented readership (e.g., non-methodological researchers,
teachers, students, government personnel, practitioners, and other professionals). Applica-
tions from a variety of disciplines such as psychology, public health, sociology, education,
and business are welcome. Books can be single- or multiple-authored or edited volumes
that: (1) demonstrate the application of a variety of multivariate methods to a single, major
area of research; (2) describe a multivariate procedure or framework that could be applied
to a number of research areas; or (3) present a variety of perspectives on a topic of interest
to applied multivariate researchers.
Anyone wishing to submit a book proposal should send the following: (1) author/title;
(2) timeline including completion date; (3) brief overview of the book’s focus, including
table of contents and, ideally, a sample chapter (or chapters); (4) a brief description of
competing publications; and (5) targeted audiences.
For more information, please contact the series editor, Lisa Harlow, at Department of
Psychology, University of Rhode Island, 10 Chafee Road, Suite 8, Kingston, RI 02881-
0808; phone (401) 874-4242; fax (401) 874-5562; or e-mail LHarlow@uri.edu.
Foreword viii
John R. Nesselroade
List of Contributors x
SECTION I
Extensions of Latent Change Score Models 7
SECTION II
Measurement and Testing Issues in
Longitudinal Modeling 143
SECTION III
Novel Applications of Multivariate
Longitudinal Methodology 261
Index 341
FOREWORD
John R. Nesselroade
become a more rigorous science and he was both innovative and enthusiastic in
the way he went about doing what he thought would make it happen. When he
had the opportunity to apply for a faculty position in the Psychology Department
at the University of Virginia, I was delighted to write a letter of recommendation
on his behalf. Parenthetically, he returned the favor 15 years later when I was
under consideration for a position at UVA. Over the years we have collaborated
on a large number of papers, chapters, and a book. As our collaboration grew, so
did our friendship. For decades now, I have been richly blessed to have him as
both dear friend and esteemed colleague.
As the reader has likely guessed by now, that young SMEP aspirant was Jack
McArdle. In the course of building his brilliant scientific career, Jack’s early hope
was realized precisely and he was indeed elected to membership in SMEP. In fact,
Jack has served SMEP as its president. Yes, Jack made it and made it big! His
enduring legacy, aimed at helping to make psychology “a quantitative, rational
science,” is attested to by the present outstanding volume. The editors and authors
have done a great mentor proud!
John R. Nesselroade
March, 2017
CONTRIBUTORS
Most phenomena in psychology and the behavioral and social sciences involve
multiple processes that unfold over time. As such, much of this research is focused
on questions related to how and why constructs change over time. Addressing these
questions requires models that can capture changes and interrelations amongst con-
structs, as well as potential mechanisms underlying such dynamics. This is not an
easy task, and is a pursuit that has kept quantitative methodologists on their toes
for many years. One of the most influential leaders in this quest for the last three
decades has been John J. (Jack) McArdle. McArdle’s work has opened the path to
models for studying longitudinal measurement, growth, and change, and set the
stage for novel approaches to examine multivariate interrelations and dynamics.
Inthisvolume,anumberofhisformerstudents,colleagues,andfriendspayatribute
to McArdle’s work, attesting to his creative thinking, his innovative methodological
advantages, and the lasting legacy of his work. In particular, the volume presents
a collection of chapters focused on the study of multivariate change. As people
develop and change, multivariate measurement of that change and analysis of those
measures can illuminate the regularities in the trajectories of individual development
as well as time dependent changes in population averages. As longitudinal data have
recently become much more prevalent in psychology and the social sciences, models
of change have become increasingly important to provide a deep understanding
of mechanisms underlying development. The chapters focus on methodological,
statistical, and modeling aspects of multivariate change, as well as applications of
longitudinal models to the study of psychological processes.
The volume is divided into three major sections, each of them dealing with a
set of methodological issues.
2 Introduction
and across participants, when this is not the case. Grimm and Jacobucci propose
using PROC NLMIXED in SAS and JAGS to estimate LCS models. Unlike stan-
dard structural equation modeling software, these programs allow the specification
of any number of time points, thus giving researchers a more fine-grained metric.
Hamagami and McArdle’s chapter presents an extension of LCS models based
on additive curvilinear effects. These effects are specified in the model together
with dynamic feedback and coupling parameters. A typical LCS specification
involves an additive latent component that is constant over time, while allow-
ing variation across individuals. In their chapter, Hamagami and McArdle extend
this specification by including an additive constant as a nonlinear function form,
and allowing the value of this latent constant to be time-dependent. For this, they
use a quadratic functional form. By incorporating the additive curvilinear effect,
the new LCS model can be used to examine dynamics in data that follow a wide
variety of nonlinear curves.
The chapter by Jacobucci and Grimm proposes regularized estimation for
examining a range of multivariate LCS models. Typically, determining the most
plausible LCS model amongst several alternative specifications involves fitting each
of the models separately and evaluating the corresponding fit statistics. Jacobucci
and Grimm’s proposed approach is a simultaneous evaluation of models based on
regularization using both maximum likelihood and Bayesian estimation. In addi-
tion to simplifying model selection, this approach allows additional parameteriza-
tions of the model, and increases the propensity of selecting the optimal model.
Boker’s chapter traces the history of path analysis and structural equation
modeling leading up to the so-called Reticular Action Model (RAM), one of
McArdle’s most influential methodological achievements. The RAM notation
allowed specifying structural equation models using three matrices only, instead
of the multiple matrices involved in standard programs at the time. This model
was implemented in software packages that allowed fitting models in a simplified
and parsimonious manner, and is still at the core of some of the most advanced
SEM programs. Boker’s chapter describes how several important technical diffi-
culties were solved in order to arrive at the RAM model. The chapter concludes
by discussing the evolution of structural equation modeling over the past 40 years.
statistic too much, leading to type I error rates below the nominal level, as well
as to a decrease in power. With the use of empirical data, the chapter illustrates
how such effects can yield spurious and even conflicting conclusions when fitting
equivalent models to the same data.
Wang and Yang’s chapter investigates the effects of over-simplifying random
components on the estimation of fixed effects in linear growth curves. Wang and
Yang use derivation work to demonstrate that, when both the random intercept
and the random slope are omitted from the specification, the standard errors of
the fixed-effects intercept and slope estimates can be biased. The authors repli-
cate their derivation findings using empirical data. The chapter concludes with
recommendations for researchers interested in fitting growth curves.
The chapter by Zhang and Liu deals with important design features for studies
that involve LCS models. Zhang and Liu carry out simulation work to determine
optimal sample size and number of measurement occasions for both univariate and
bivariate LCS models. They propose a method that allows researchers to calculate
the power for estimating individual model parameters using their own R statistical
package RAMpath.
Hayes’s chapter reports analyses examining the performance of two strategies
for dealing with missing data in longitudinal studies. Hayes compares CART (clas-
sification and regression trees) and random forest analyses when such techniques
are applied to multiple imputation and when they are used to model the proba-
bility of dropout through the creation of inverse probability weights. These two
approaches are evaluated under incomplete data due to “missing not at random”
(MNAR) mechanisms with small sample sizes. Hayes’s findings include that, in
small samples, CART and weighting methods outperform multiple imputation,
indicating that these approaches can help to deal with missing data problems in
empirical research.
In his chapter, Estabrook proposes a longitudinal perspective for measurement
evaluation, suggesting an extension of a measurement model based on derivatives.
This proposed approach builds on McArdle’s factor of curves (FOCUS) model,
in which the intercepts and slopes corresponding to growth curves of multiple
observed variables (or items) repeated over time are factor analyzed, yielding a
factor of the intercepts and a factor of the slopes. Estabrook’s proposed approach
provides a useful framework to address issues related to measurement evaluation
over time. This framework facilitates the integration of within-person change and
intra-individual processes into behavioral research.
and validity, growth and dynamics, and provide examples of current research on
which McArdle has left his indelibly mark, such as behavioral genetics, research
in collegiate athletics, and studies of minorities.
The chapter by Bowles focuses on measurement in developmental longitudinal
modeling. Bowles demonstrates how longitudinal studies are sensitive to violations
of interval level measurement, arguing that such violations have important impli-
cations regarding validity. Bowles describes three approaches to provide evidence
of interval scaling: mechanistic modeling, latent variable modeling, and Rasch
measurement. After discussing the strengths and limitations of each approach, the
chapter concludes with recommendations for empirical research.
Ram and Yang continue the theme of developmental methodology. In their
chapter, they focus on models for examining dynamics in developmental stud-
ies, for which they use techniques implemented in fisheries management. At
the core of their proposal is the integration of inter-individual differences with
intra-individual change. For this, they argue for a switch from the integrated
mathematical forms commonly used in growth modeling to differential equations.
Such equations, Ram and Yang maintain, allow a richer theoretical description
of the model and provide a more cogent articulation of the actual processes
underlying the change.
The chapter by Neale traces important developments in the genetic studies
of growth and development. Neale emphasizes the contributions to this research
by McArdle, including early developments in the psychometric and biometric
models, and more recent work encompassing multivariate latent growth curves
for twin data. In addition to reviewing these methodological developments, the
chapter describes potential extensions and applications of such models.
Petr and Paskus’s chapter depicts McArdle’s work for the National Collegiate
Athletic Association (NCAA). They narrate how, over 25 years, McArdle’s work
set the stage for high-level research at the NCAA that focused on how schools
can predict and promote the academic and life successes of college athletes. Petr
and Paskus describe how this work has changed, not just the research conducted
at the NCAA, but the very landscape of college athletics.
The final chapter in the volume, by Hishinuma and colleagues, illustrates
another important contribution by McArdle to current research, namely his work
on multivariate statistics as applied to research on minorities. Hishinuma and col-
leagues describe McArdle’s interactions with the personnel at the University of
Hawai’i at Mānoa and emphasize the profound impact of such interactions, not
only on the team but on the research area at large. The chapter includes personal
notes depicting McArdle’s rare combination of skills that allows him to blend into
and become part of an organization’s culture. These contributions, Hishinuma
and colleagues tell us, have been invaluable and have set the foundation for all
future endeavors of the research team.
In sum, we are pleased to present Longitudinal Multivariate Psychology. As its
title indicates, this volume is intended to offer a window into state-of-the-art
6 Introduction
Extensions of Latent
Change Score Models
1
METHODOLOGICAL ISSUES AND
EXTENSIONS TO THE LATENT
DIFFERENCE SCORE FRAMEWORK1
Linying Ji & Sy-Miin Chow
Few topics in the social and behavioral sciences are as fundamental and as riddled
with controversies as the study of change. From the questions of whether and
how intervention programs impact the targeted behaviors of interest, to studies
of lifespan developmental changes, the pursuit of change arises in a variety of
contexts and brings its own fair share of methodological challenges. Particularly
well-known in the history of studying change are the various disputes surrounding
the analysis of observed difference scores (see e.g., Bereiter, 1963; Cattell, 1966;
Cronbach & Furby, 1970; Harris, 1963; Lord, 1956, 1958; Nesselroade & Cable,
1974), such as the low reliability of observed difference scores, and difficulties in
interpreting results from factor analyzing such scores. One of the key contribu-
tions of McArdle to the fields of psychometrics and lifespan development resides
in his proposal to formalize modeling of change scores at the latent variable level.
Among the latent difference models proposed by McArdle and colleagues, is the
univariate dual change score model, which allows researchers to examine data
manifesting nonlinear (specifically, sigmoid-shaped) curves as latent difference
score (LDS) models with linear change functions (McArdle & Hamagami, 2001).
LDS models allow researchers to capture change at the latent level, which
remedies the low reliability of change scores related directly to the measurement
errors of the testing tools. In addition, formulating change hypotheses within a
latent variable modeling framework also expands the ways in which researchers
may examine the determinants or pathways of intraindividual change and com-
plex interdependencies among multiple change processes – some of the rationales
for conducting longitudinal research proposed by Baltes and Nesselroade (1979).
Such work builds on the extensive literature on dynamic models in the form of
differential or difference equations.
Differential equations are used to represent a system of variables that evolve
continuously over time, whereas difference equation model are designed for
10 Linying Ji & Sy-Miin Chow
framework by several researchers (du Toit & Browne, 2007; Hamaker, 2005), is the
“model-implied” approach for partially stable processes. This approach assumes
stability2 in the portion of a modeled process that shows lagged dependencies,
which, in turn, requires all associated modeling parameters to be time invari-
ant. Imposing such assumptions then enables researchers to use the unconditional
mean and covariance matrix implied by the model to specify the initial distribu-
tion on the latent variables at the first observed time point. du Toit and Browne
(2007) showed that if the true underlying processes do in fact conform to a par-
tially stable structure and are well described by the “model implied” approach,
the IC structure can be recovered well using the “free-parameter” approach. We
thus do not discuss the “model-implied” approach further here.
Another IC specification commonly adopted in the state-space literature is a
diffuse IC specification in which the initial mean vector of the latent variables, μ1 ,
is set to a vector of zeros; and the corresponding initial latent variable covariance
matrix, 1 , is specified to be a diagonal matrix with variance entries fixed at arbi-
trarily large values (e.g., De Jong, 1991; Harvey, 2001; Harvey & Phillips, 1979;
Schweppe, 1973). Larger values of variances reflect greater uncertainties in the
structure of the IC. This IC specification assumes that limited information is avail-
able to specify the IC structure with high precision. Thus, the means of the initial
latent variables are all set to zero – a predicted value that is relatively reasonable
for many stable, time-intensive longitudinal processes (i.e., time series processes)
frequently considered in the econometric literature in which the diffuse IC speci-
fication is adopted. With the large numbers of time points typically available from
those studies, such misspecification of IC has relatively trivial consequences. This
is not the case, however, when panel data with a limited number of occasions is
involved (Chow et al., 2016; Losardo, 2012; Oud et al., 1990; Ou et al., 2016).
Indeed, the issue of IC specification has received considerable attention in the
state-space framework (De Jong, 1991; Harvey, 2001) and much can be learned
from some of the issues, problems, and solutions proposed in that literature. Pre-
vious studies have investigated the consequences of misspecification of IC in the
context of a vector autoregressive process with moving average residuals (du Toit
& Browne, 2007), an autoregressive latent trajectory model (Ou, Chow, Ji, &
Molenaar, 2016),and with both stable and unstable processes (Losardo, 2012).
However, the effects of IC misspecification have not been evaluated within the
context of the BDCS model. Our emphasis in this chapter is to evaluate whether
and to what extent misspecification in IC may interact with features of the differ-
ence equations implicated in the BDCS model. In particular, we examine whether
the BDCS model can reliably distinguish among variations due to interindividual
differences in IC, measurement errors, and process noises when model estimation
is performed using panel data with limited information.
As an initial demonstration of the consequences of IC misspecification in
the BDCS model in the absence of process noise, we simulated a set of data
using the BDCS model in Equations 1.1, 1.4, and 1.5; with N = 500 and two
configurations of total number of time points (T = 5 and T = 10). We then fit the
Extensions to LDS Framework 15
BDCS model with the correctly specified IC structure, and the diffuse IC speci-
fication with all variance terms in 1 set to a sequence of possible values: .01, 1,
10, and 100. Table 1.1 summarizes the true values of the parameters used for data
generation purposes, as well as key simulation results based on fitting variations of
the BDCS model with freely estimated IC (the correctly specified IC), and mis-
specified IC specifications with different levels of “diffuseness.” We also show in
Table 1.1: (1) the total bias of the parameter estimates (over nine possible param-
eters), defined as the sum of the difference between each parameter’s estimated
and true values across all modeling parameters; and (2) the root mean squared
error (RMSE), defined as the square root of the mean of the squared differences
between the true and estimated values across all modeling parameters. In addition,
we also report the Akaiki Information Criterion (AIC) and Bayesian Information
Criteria (BIC) as model selection measures. We found that differences among the
conditions where the variances in 1 were set to 1, 10, and 100 were minimal.
Thus, we only included results from the condition with IC variances fixed at 10
due to space constraints.
As shown in Table 1.1, with T = 5 and the correctly specified IC, all of the
model parameters were recovered very well, with a small total bias and RMSE.
Both the AIC and the BIC selected the correctly specified IC structure over the
other IC structures. Of the conditions with diffuse IC specification, when the IC
covariance matrix was set to 1 = 10×I4 , only a slight increase in total bias was
observed. However, there was a five-fold increase in RMSE. When the number of
time points was increased from 5 to 10, all parameter estimates were very close to
their true values. The total bias and RMSE from the diffuse IC with 1 = 10 × I4
and T = 10 were close to those observed in the correctly specified IC. Also, as
expected, with larger T, smaller values of standard error estimates were observed.
The level of diffuseness of the IC also affected the quality of the point estimates.
In general, if the variances of the IC were specified to be large enough for the
misspecified IC distribution to show a reasonable amount of overlap with the true
IC distribution, estimation results were minimally affected. On the contrary, if the
variances of the IC were specified to be too restrictive – for instance, when some
of the interindividual variance terms were incorrectly fixed at 0, considerable
decrements in point estimates are noted. As an example, when we decreased the
IC variances to .01, all parameter estimates showed substantial biases and RMSEs,
even with 10 measurement occasions (see Table 1.1). The AIC and the BIC,
again, preferred models with more diffuse (e.g., 1 = 10 × I4 ) than the model
with the least diffuse but misspecified IC (1 = 0.01 × I4 ).
It should be noted that a sheer increase in the number of time points, per
se, may not always be adequate in overcoming issues pertaining to the misspec-
ification of the IC. For instance, when the absolute values of the self-feedback
parameters (i.e., the βs) are large, the changes in the latent processes (differences
in the latent processes between successive time points) would be larger and the
processes would reach their asymptotic levels faster. As shown in Figure 1.1, when
β1 = −.3 (the two plots in the top panel), the latent processes do not reach their
Another random document with
no related content on Scribd:
The Project Gutenberg eBook of
Menschenwee
This ebook is for the use of anyone anywhere in the United
States and most other parts of the world at no cost and with
almost no restrictions whatsoever. You may copy it, give it away
or re-use it under the terms of the Project Gutenberg License
included with this ebook or online at www.gutenberg.org. If you
are not located in the United States, you will have to check the
laws of the country where you are located before using this
eBook.
Title: Menschenwee
Roman van het land
Language: Dutch
MENSCHENWEE
[Inhoud]
MENSCHENWEE
DOOR
IS. QUÉRIDO
HAARLEM
DE ERVEN F. BOHN
Eerste Boek
WINTER.
[1]
[Inhoud]
EERSTE HOOFDSTUK.
[Inhoud]
I.
Sai vloekte en skolde.… main f’r lailikke fint.… en wèg woei weer,
beverig-zwak Piet’s zeurig geluid in den wind over de
akkersombering vergàlmend.
Met zware dubbele hypotheek belast, was het bunder land van den
ouën Gerrit, waarop z’n zoons nu werkten. Maar voor hem gelukkig,
lag ’t in het hartje van ’t tuinderstedeke Wiereland.—Z’n anderen
grond had ie gepacht, lag meer aan den kant van gemeente
Duinkijk, die komvormig om Wiereland liep en in één weg zich
uitstrekte recht-door naar zee, vormend daar één gemeente met
Zeekijk, armelijk zeegehucht, dat alleen zomers, door kleurige
vreugde van badgasten wat opleefde. Zijn bunder werd nou ’t laatst
bewerkt. Achter z’n huisje en erf liep ’t uit, omsloten rechts, door
groezelige armelijke huisachterbuurtjes uit de arbeiders- en
tuinderswijk, links openliggend, met aanschakeling van tuinders- en
bollenakkers, ver, heel ver.—Dooie tijd was ’t alom voor de tuinders
in ’t stedeke; luiertijd en tijd van ellende voor de meesten, pijnigende
zorg-armoe en grauw gebrek voor de Wierelandsche getrouwde
daggelders, los volk, dat met ’t sterven van den [4]zomer, werkeloos
rondscharrelde; hier ’n trap, daar ’n snauw, ’n paar dagen zwaar
spittend, dan weer niks, geradbraakt en omhongerend elken
winterdag, in gierig licht en zonnestraalte. De ouë Gerrit had z’n
twee pootige zoons, die hem te eten gaven, nou hij ’t jaren hun had
gedaan. Want nou, op z’n ouën dag, voelde ie zich na z’n zware
ziekte van verleden jaar te stram en te beroerd om in guurte en
regen te spitten. Als ie z’n kinderen niet had, zou ie nou nog op z’n
achtenzestigste jaar moèten, want loon kon ie niet betalen.
Lekker warmde ’t nou in ’m, dat hij maar met z’n handen in de
zakken daar te staan had, alleen maar kijken naar hun werk en ie z’n
eigen niet druk hoefde te maken. Wat ’n kerel was z’n Dirk toch!.…
hoho!.… wat ’n skoften.… wat ’n makelei.… wat ’n poote en
arreme!.… Jesis!.… wa stong ie daar stoer in z’n kiel.… huhu!.…
hu.… da skept moar, .… skiet moar òver. En s’n Piet gunters.… ook
’n pootige knoest.… kaik.… da song ie weer.… wat ’n hassebas.…
da se’m toch soo gemeen konne f’rbluffe.… kaik die Dirk.… da segt
t’met g’n woord.… stil.… da gong ie puur.… bònk!.… poef.… bònk!..
poef!! wat ’n vracht.… daa’s werk.… ieder keer t’ met vaiftig pond.…
boofe se macht.. main kristis.… Dâ ken puur nie eene van de
ploas!.…..
Juist wou Dirk z’n graaf weer in den grond steken, toen ie ouen Bolk
van het Wierelandsche achtereind, ’t droefste, afgekniesdste
ellendehoekje van de streek, naar hen toe zag dobberen.